Exact Significance Testing with Biased Coin Randomization

摘要

For a clinical trial where two treatments have been assigned sequentially to patients via Efron's (1971) biased coin design, a recursion procedure is derived for obtaining the exact randomization distribution of a class of test statistics. This enables one to perform exact significance tests of the hypothesis of no treatment difference. The randomization distribution of the statistic is conditional on the imbalance of the treatment allocation. It is illustrated that if the analysis is performed as if complete randomization was used, conservative and anticonservative errors can be incurred. The applicability of the test to censored data is also discussed. (Author)